1
# Copyright (C) 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21
from bzrlib.deprecated_graph import (node_distances, select_farthest)
22
from bzrlib.revision import NULL_REVISION
24
# DIAGRAM of terminology
34
# In this diagram, relative to G and H:
35
# A, B, C, D, E are common ancestors.
36
# C, D and E are border ancestors, because each has a non-common descendant.
37
# D and E are least common ancestors because none of their descendants are
39
# C is not a least common ancestor because its descendant, E, is a common
42
# The find_unique_lca algorithm will pick A in two steps:
43
# 1. find_lca('G', 'H') => ['D', 'E']
44
# 2. Since len(['D', 'E']) > 1, find_lca('D', 'E') => ['A']
48
class _StackedParentsProvider(object):
50
def __init__(self, parent_providers):
51
self._parent_providers = parent_providers
54
return "_StackedParentsProvider(%r)" % self._parent_providers
56
def get_parents(self, revision_ids):
57
"""Find revision ids of the parents of a list of revisions
59
A list is returned of the same length as the input. Each entry
60
is a list of parent ids for the corresponding input revision.
62
[NULL_REVISION] is used as the parent of the first user-committed
63
revision. Its parent list is empty.
65
If the revision is not present (i.e. a ghost), None is used in place
66
of the list of parents.
69
for parents_provider in self._parent_providers:
70
pending_revisions = [r for r in revision_ids if r not in found]
71
parent_list = parents_provider.get_parents(pending_revisions)
72
new_found = dict((k, v) for k, v in zip(pending_revisions,
73
parent_list) if v is not None)
74
found.update(new_found)
75
if len(found) == len(revision_ids):
77
return [found.get(r, None) for r in revision_ids]
81
"""Provide incremental access to revision graphs.
83
This is the generic implementation; it is intended to be subclassed to
84
specialize it for other repository types.
87
def __init__(self, parents_provider):
88
"""Construct a Graph that uses several graphs as its input
90
This should not normally be invoked directly, because there may be
91
specialized implementations for particular repository types. See
92
Repository.get_graph()
94
:param parents_func: an object providing a get_parents call
95
conforming to the behavior of StackedParentsProvider.get_parents
97
self.get_parents = parents_provider.get_parents
98
self._parents_provider = parents_provider
101
return 'Graph(%r)' % self._parents_provider
103
def find_lca(self, *revisions):
104
"""Determine the lowest common ancestors of the provided revisions
106
A lowest common ancestor is a common ancestor none of whose
107
descendants are common ancestors. In graphs, unlike trees, there may
108
be multiple lowest common ancestors.
110
This algorithm has two phases. Phase 1 identifies border ancestors,
111
and phase 2 filters border ancestors to determine lowest common
114
In phase 1, border ancestors are identified, using a breadth-first
115
search starting at the bottom of the graph. Searches are stopped
116
whenever a node or one of its descendants is determined to be common
118
In phase 2, the border ancestors are filtered to find the least
119
common ancestors. This is done by searching the ancestries of each
122
Phase 2 is perfomed on the principle that a border ancestor that is
123
not an ancestor of any other border ancestor is a least common
126
Searches are stopped when they find a node that is determined to be a
127
common ancestor of all border ancestors, because this shows that it
128
cannot be a descendant of any border ancestor.
130
The scaling of this operation should be proportional to
131
1. The number of uncommon ancestors
132
2. The number of border ancestors
133
3. The length of the shortest path between a border ancestor and an
134
ancestor of all border ancestors.
136
border_common, common, sides = self._find_border_ancestors(revisions)
137
return self._filter_candidate_lca(border_common)
139
def find_difference(self, left_revision, right_revision):
140
"""Determine the graph difference between two revisions"""
141
border, common, (left, right) = self._find_border_ancestors(
142
[left_revision, right_revision])
143
return (left.difference(right).difference(common),
144
right.difference(left).difference(common))
146
def _make_breadth_first_searcher(self, revisions):
147
return _BreadthFirstSearcher(revisions, self)
149
def _find_border_ancestors(self, revisions):
150
"""Find common ancestors with at least one uncommon descendant.
152
Border ancestors are identified using a breadth-first
153
search starting at the bottom of the graph. Searches are stopped
154
whenever a node or one of its descendants is determined to be common.
156
This will scale with the number of uncommon ancestors.
158
As well as the border ancestors, a set of seen common ancestors and a
159
list of sets of seen ancestors for each input revision is returned.
160
This allows calculation of graph difference from the results of this
163
if None in revisions:
164
raise errors.InvalidRevisionId(None, self)
165
common_searcher = self._make_breadth_first_searcher([])
166
common_ancestors = set()
167
searchers = [self._make_breadth_first_searcher([r])
169
active_searchers = searchers[:]
170
border_ancestors = set()
171
def update_common(searcher, revisions):
172
w_seen_ancestors = searcher.find_seen_ancestors(
174
stopped = searcher.stop_searching_any(w_seen_ancestors)
175
common_ancestors.update(w_seen_ancestors)
176
common_searcher.start_searching(stopped)
179
if len(active_searchers) == 0:
180
return border_ancestors, common_ancestors, [s.seen for s in
183
new_common = common_searcher.next()
184
common_ancestors.update(new_common)
185
except StopIteration:
188
for searcher in active_searchers:
189
for revision in new_common.intersection(searcher.seen):
190
update_common(searcher, revision)
193
new_active_searchers = []
194
for searcher in active_searchers:
196
newly_seen.update(searcher.next())
197
except StopIteration:
200
new_active_searchers.append(searcher)
201
active_searchers = new_active_searchers
202
for revision in newly_seen:
203
if revision in common_ancestors:
204
for searcher in searchers:
205
update_common(searcher, revision)
207
for searcher in searchers:
208
if revision not in searcher.seen:
211
border_ancestors.add(revision)
212
for searcher in searchers:
213
update_common(searcher, revision)
215
def _filter_candidate_lca(self, candidate_lca):
216
"""Remove candidates which are ancestors of other candidates.
218
This is done by searching the ancestries of each border ancestor. It
219
is perfomed on the principle that a border ancestor that is not an
220
ancestor of any other border ancestor is a lowest common ancestor.
222
Searches are stopped when they find a node that is determined to be a
223
common ancestor of all border ancestors, because this shows that it
224
cannot be a descendant of any border ancestor.
226
This will scale with the number of candidate ancestors and the length
227
of the shortest path from a candidate to an ancestor common to all
230
searchers = dict((c, self._make_breadth_first_searcher([c]))
231
for c in candidate_lca)
232
active_searchers = dict(searchers)
233
# skip over the actual candidate for each searcher
234
for searcher in active_searchers.itervalues():
236
while len(active_searchers) > 0:
237
for candidate, searcher in list(active_searchers.iteritems()):
239
ancestors = searcher.next()
240
except StopIteration:
241
del active_searchers[candidate]
243
for ancestor in ancestors:
244
if ancestor in candidate_lca:
245
candidate_lca.remove(ancestor)
246
del searchers[ancestor]
247
if ancestor in active_searchers:
248
del active_searchers[ancestor]
249
for searcher in searchers.itervalues():
250
if ancestor not in searcher.seen:
253
# if this revision was seen by all searchers, then it
254
# is a descendant of all candidates, so we can stop
255
# searching it, and any seen ancestors
256
for searcher in searchers.itervalues():
258
searcher.find_seen_ancestors(ancestor)
259
searcher.stop_searching_any(seen_ancestors)
262
def find_unique_lca(self, left_revision, right_revision):
263
"""Find a unique LCA.
265
Find lowest common ancestors. If there is no unique common
266
ancestor, find the lowest common ancestors of those ancestors.
268
Iteration stops when a unique lowest common ancestor is found.
269
The graph origin is necessarily a unique lowest common ancestor.
271
Note that None is not an acceptable substitute for NULL_REVISION.
272
in the input for this method.
274
revisions = [left_revision, right_revision]
276
lca = self.find_lca(*revisions)
280
raise errors.NoCommonAncestor(left_revision, right_revision)
283
def iter_topo_order(self, revisions):
284
"""Iterate through the input revisions in topological order.
286
This sorting only ensures that parents come before their children.
287
An ancestor may sort after a descendant if the relationship is not
288
visible in the supplied list of revisions.
290
sorter = tsort.TopoSorter(zip(revisions, self.get_parents(revisions)))
291
return sorter.iter_topo_order()
293
def is_ancestor(self, candidate_ancestor, candidate_descendant):
294
"""Determine whether a revision is an ancestor of another.
296
There are two possible outcomes: True and False, but there are three
297
possible relationships:
299
a) candidate_ancestor is an ancestor of candidate_descendant
300
b) candidate_ancestor is an descendant of candidate_descendant
301
c) candidate_ancestor is an sibling of candidate_descendant
303
To check for a, we walk from candidate_descendant, looking for
306
To check for b, we walk from candidate_ancestor, looking for
307
candidate_descendant.
309
To make a and b more efficient, we can stop any searches that hit
312
If we exhaust our searches, but neither a or b is true, then c is true.
314
In order to find c efficiently, we must avoid searching from
315
candidate_descendant or candidate_ancestor into common ancestors. But
316
if we don't search common ancestors at all, we won't know if we hit
317
common ancestors. So we have a walker for common ancestors. Note that
318
its searches are not required to terminate in order to determine c to
321
ancestor_walker = self._make_breadth_first_searcher(
322
[candidate_ancestor])
323
descendant_walker = self._make_breadth_first_searcher(
324
[candidate_descendant])
325
common_walker = self._make_breadth_first_searcher([])
326
active_ancestor = True
327
active_descendant = True
328
while (active_ancestor or active_descendant):
330
if active_descendant:
332
nodes = descendant_walker.next()
333
except StopIteration:
334
active_descendant = False
336
if candidate_ancestor in nodes:
338
new_common.update(nodes.intersection(ancestor_walker.seen))
341
nodes = ancestor_walker.next()
342
except StopIteration:
343
active_ancestor = False
345
if candidate_descendant in nodes:
347
new_common.update(nodes.intersection(
348
descendant_walker.seen))
350
new_common.update(common_walker.next())
351
except StopIteration:
353
for walker in (ancestor_walker, descendant_walker):
354
for node in new_common:
355
c_ancestors = walker.find_seen_ancestors(node)
356
walker.stop_searching_any(c_ancestors)
357
common_walker.start_searching(new_common)
361
class _BreadthFirstSearcher(object):
362
"""Parallel search the breadth-first the ancestry of revisions.
364
This class implements the iterator protocol, but additionally
365
1. provides a set of seen ancestors, and
366
2. allows some ancestries to be unsearched, via stop_searching_any
369
def __init__(self, revisions, parents_provider):
370
self._start = set(revisions)
371
self._search_revisions = None
372
self.seen = set(revisions)
373
self._parents_provider = parents_provider
376
return ('_BreadthFirstSearcher(self._search_revisions=%r,'
377
' self.seen=%r)' % (self._search_revisions, self.seen))
380
"""Return the next ancestors of this revision.
382
Ancestors are returned in the order they are seen in a breadth-first
383
traversal. No ancestor will be returned more than once.
385
if self._search_revisions is None:
386
self._search_revisions = self._start
388
new_search_revisions = set()
389
for parents in self._parents_provider.get_parents(
390
self._search_revisions):
393
new_search_revisions.update(p for p in parents if
395
self._search_revisions = new_search_revisions
396
if len(self._search_revisions) == 0:
397
raise StopIteration()
398
self.seen.update(self._search_revisions)
399
return self._search_revisions
404
def find_seen_ancestors(self, revision):
405
"""Find ancestors of this revision that have already been seen."""
406
searcher = _BreadthFirstSearcher([revision], self._parents_provider)
407
seen_ancestors = set()
408
for ancestors in searcher:
409
for ancestor in ancestors:
410
if ancestor not in self.seen:
411
searcher.stop_searching_any([ancestor])
413
seen_ancestors.add(ancestor)
414
return seen_ancestors
416
def stop_searching_any(self, revisions):
418
Remove any of the specified revisions from the search list.
420
None of the specified revisions are required to be present in the
421
search list. In this case, the call is a no-op.
423
stopped = self._search_revisions.intersection(revisions)
424
self._search_revisions = self._search_revisions.difference(revisions)
427
def start_searching(self, revisions):
428
if self._search_revisions is None:
429
self._start = set(revisions)
431
self._search_revisions.update(revisions.difference(self.seen))
432
self.seen.update(revisions)
added
removed
bzrlib/graph.py
